Convolutional neural network using magnetic resonance brain imaging to predict outcome from tuberculosis meningitis

Abstract

Abstract Introduction Tuberculous meningitis (TBM) leads to high mortality, especially amongst individuals with HIV. Predicting the incidence of disease-related complications is challenging, for which purpose the value of brain magnetic resonance imaging (MRI) has not been well investigated. We used a convolutional neural network (CNN) to explore the contribution of brain MRI to the conventional prognostic determinants. Method We used data from two randomised control trials of HIV-positive and HIV-negative adults with clinical TBM in Vietnam to predict the occurrence of death or new neurological complications in the first two months after the subject’s first MR session. We developed and compared three models: a logistic regression with no imaging data as reference, a CNN that utilised only T1-weighted MR volumes, and a model that fused both. All models were fine-tuned using two repeated 5-fold cross-validation. The final evaluation was based on a random 70/30 training/test split, stratified by the outcome and HIV status. Based on the selected model, we explored the interpretability maps derived from the models. Results 215 patients were included, with an event prevalence of 22.3%. On the test set our non-imaging model had higher AUC (71.2% ± 1.1%) than the imaging-only model (67.3% ± 2.6%). The fused model was superior to both, with an average AUC = 77.3% ± 4.0% in the test set. The non-imaging variables were more informative in the HIV-positive group, while the imaging features were more predictive in the HIV-negative group. All three models performed better in the latter cohort. Across 10 folds, all models’ performance varied considerably, suggesting room for improvement. The interpretability maps show the model’s focus on the lateral fissures, the corpus callosum and the tissue around lateral ventricles. Some standard lesions such as tuberculomas and basal enhancement, were not determined by the model to be relevant disease prognostic features. Conclusion Imaging information using a CNN can help predict unwanted outcomes of TBM. A larger dataset is now needed to confirm our findings.